用于3维重建的立体图像校正

传统的图像校正方法通过最小化匹配点间的距离作为目标函数引导图像校正,但往往会导致被校正图像出现较大的失真变形,不仅无法为3维(3D)重建提供信息,反而会带来额外的误差.为此提出一种无需相机标定的图像校正方法,有效地保留图像因拍摄方位不同带来的视差,为3D重建提供所需的重要信息.方法首先利用立体图像的对极匹配原理,推导出将2幅图像极线对齐到同一方向的变换,然后在此基础上引入使得图像纵横比保持不变的剪切变换过程,有效地减小校正图像的几何变形.实验表明,该方法是一种鲁棒的图像校正方法.

Stereo Image Rectification Suitable for 3D Reconstruction

Traditional methods of image rectification aim to minimizing the least-squares distance between the matching points, however, the rectification process may cause great geometric distortion, which not only makes it impossible to offer information for 3-dimensional reconstruction, but also results in extra errors. In this paper, an improved algorithm proposed which minimize the distortion by combining a newly developed projective transform with a properly chosen shearing transform. Firstly, projective transformation is developed to make the polar lines of two images align to the same direction. Secondly, shear transformation is introduced to ensure the image aspect ratio remain constant. Experimental result demonstrate that our proposed method is quite robust and is well qualified to the rectification for 3-D reconstruction.